A failure-type specific risk prediction tool for selection of head-and-neck cancer patients for experimental...

The objective of this work was to develop a tool for decision support, providing simultaneous predictions of the risk of loco-regional failure (LRF) and distant metastasis (DM) after definitive treatment for head-and-neck squamous cell carcinoma (HNSCC).

Retrospective data for 560HNSCC patients were used to generate a multi-endpoint model, combining three cause-specific Cox models (LRF, DM and death with no evidence of disease (death NED)). The model was used to generate risk profiles of patients eligible for/included in a de-intensification study (RTOG 1016) and a dose escalation study (CONTRAST), respectively, to illustrate model predictions versus classic inclusion/exclusion criteria for clinical trials. The model is published as an on-line interactive tool (https://katrin.shinyapps.io/HNSCCmodel/).

The final model included pre-selected clinical variables (tumor subsite, T stage, N stage, smoking status, age and performance status) and one additional variable (tumor volume). The treatment failure discrimination ability of the developed model was superior of that of UICC staging, 8^th edition (AUC_LRF=72.7% vs 64.2%, p<0.001 and AUC_DM=70.7% vs 58.8%, p<0.001). Using the model for trial inclusion simulation, it was found that 14% of patients eligible for the de-intensification study had>20% risk of tumor relapse. Conversely, 9 of the 15 dose escalation trial participants had LRF risks<20%.

A multi-endpoint model was generated and published as an on-line interactive tool. Its potential in decision support was illustrated by generating risk profiles for patients eligible for/included in clinical trials for HNSCC.